Integrating data from multidisciplinary Management of Malignant Pleural Mesothelioma: a cohort study

The PDTA (Percorso Diagnostico Terapeutico e Assistenziale-Territorial Diagnostic, Therapeutic and Assistance Planning) for MPM was first defined in 2014 within the Provincial Oncology Intercompany Department (DIPO) of Pavia. In 2018, the PDTA established two territorial outpatient facilities in view of local epidemiological data on asbestos-related diseases: a first-level clinic, located at the PRESST in the town of Broni (in the province of Pavia), and a second-level clinic managed by the Pneumology Unit and Medical Oncology Unit at the IRCCS Policlinico San Matteo (University hospital, located in Pavia). The two structures cooperate to provide care for patients with suspected asbestos-related pathology. They perform first-level diagnostic investigations as well as those of greater complexity to allow diagnosis and tumor staging. Moreover, they cooperate to provide an individualized therapeutic pathway and psycho-social care plan for the patients. In addition, the PDTA manages the chemotherapy treatments available at our territorial facilities (PRESST Broni, Broni-Stradella Hospital), referring to IRCCS San Matteo when high complexity investigations are needed.

Since November 2018, we have been collecting data on all mesothelioma patients managed through the PDTA. The number of patients was then narrowed by several criteria to be included in the study, as represented in Fig. 1. Data were obtained from the anamnestic records, outpatient reports and hospital discharge letters. Informed consent of each patients was collected routinely at hospital admission in accordance with standard hospital procedures. Imaging findings were graded by two radiologists in blind; the experienced radiologist was the referring radiologist for the “multidisciplinary board of thoracic neoplasms” while the non-experienced radiologist was a radiology resident with less than 5-year-experience in thoracic radiology. Disagreement between the two radiologists were solved by mutual agreement.

Data collection was made in the form of Excel Spreadsheets. Basic statistical analysis was conducted through the Excel “Data Analysis” add-on package. Survival curves were estimated with Kaplan-Meier method. Advanced statistical analysis of data was performed by using the JMP partition algorithm (JMP-Statistical Discoveries. From SAS, website at www.​jmp.​com) which is able to search all possible splits of best response predictors. Decision trees are a classification methodology, wherein the classification process is constructed with the use of a set of hierarchical decisions on the feature variables, arranged in a tree-like structure. The decision at a particular node of the tree, which is referred to as the split criterion, is typically a condition on one or more feature variables in the training data. These splits (or partitions) of the data are done recursively in top-down fashion by using the split criteria, through an iterative approach. A small number of features are removed in each iteration, according to the split criteria. Then, the classifier is retrained on the pruned set of features to re-estimate the weights. The decision tree is typically constructed as a hierarchical partitioning of the training examples, just as a top-down clustering algorithm partitions the data hierarchically. The main difference from clustering is that the partitioning criterion in the decision tree is supervised with the class label in the training instances. Partitioning is simple to implement and highly interpretable and a useful application is to create a diagnostic heuristic for a disease. Given symptoms and outcomes for a population, partitioning can be used to generate a hierarchy of questions to help diagnose new patients. Predictors (split-criteria) can be either continuous or categorical (nominal or ordinal). If a predictor is continuous, then the splits are created by a cutting value. Thus, the sample is divided into values below and above the cutting one. If a predictor is categorical, then the sample is divided into two groups of levels: continuous or categorical (nominal or ordinal). If the response is continuous, then the platform fits the means of the response values. If the response is categorical, then the fitted value is a probability for the levels of the response. To properly estimate the residual uncertainty of classification, several numerical indexes are used. Among them the Gini heterogeneity index (I_G) is calculable from relative frequencies (p_i) of each of the M total classes [14]. Another one is the entropy (H) which is associated with the concept of quantity of uncertainty [15]. By using one of the above-mentioned indexes the algorithm is able to make a decision on which partition to make at each step of tree construction.

We further analyzed the records through partition analysis to explore a combination of factors that could impact clinical outcome, to identify and select the most relevant predictive and prognostic variables. Differences emerged by subdividing the cohort based on patient gender. We thus focused on the most relevant outcome parameter, namely overall survival. We expected that the high number of early-stage diseases in our population study would correlate with a relatively better prognosis if compared to advanced ones, but, quite surprisingly, no statistically significant association could be identified between OS and TNM stage at diagnosis (Fig. 3, panel I). The most relevant variable associated to OS in women was determined by access to second line treatment, since those patients who did not receive it displayed a significantly lower OS (12.5 months vs 22.4 months, Fig. 3, panel II-C). Among patients who received second line therapy, the subsequent split underlined that exposure to cigarette smoke (past and/or current) significantly affected mortality (average OS 18.7 months). Although, performance status (PS) should be a limitation in defining therapeutic strategies, these data suggested that, at least in females who never smoked, comprehensive chemotherapeutic regimens can assure better outcomes (mean OS 25.6 months). Concerning males, it should be noted that no significant splits could be found when patients were stratified by age. When removing this variable from the analysis (Fig. 3, panel II-D), chemotherapy schedule significantly impacted OS and treatment with platinum and pemetrexed was associated with OS rates (15.4 months) that were higher than treatment with platinum alone (11.4 months). Coherently, concomitant advanced disease stage was associated with worse prognosis (mean OS of 5 months). Again, although PS and comorbidities might guide the decision between different treatment schedules, our study demonstrated that such decisions were irrelevant and all patients without contraindication should receive dual chemotherapy regimens. It should be noted that the results of the analysis impacted long term overall outcome and no significant differences could be found by evaluating chemotherapy regimen and progression free survival (PSF) in both men and female patients (Fig. 3, panel III). Since no clear data are available in the literature on the most beneficial conventional chemotherapy agent [30, 31], we proceeded to analyze data regarding second line treatments. Specifically, we compared the efficacy of gemcitabine (30 patients) vs vinorelbine (19 patients). We excluded from the analysis the three patients who underwent first line chemotherapy re-challenge. Based on distribution comparison a statistically significant different has been observed in OS in female patients treated with vinorelbine in second line setting (Fig. 3, panel IV-G) whereas in the male subgroup the Student’ t test comparing the pair did not reach a statistical significance. Within the limit of the cohort in analysis, this finding could be an interesting open point requiring deep investigation in dedicated works in next future.